Image Processing & Applications

The extraction of quantitative parameters (features) from medical images to build prediction models for computer-assisted clinical decision making (radiomics) has attracted a lot of research attention in recent years. Radiomics features may be influenced by the size of the region of interest (ROI) and it is recommended to employ volume-independent features to reduce the risk of biased prediction models. Despite being a crucial factor, the influence of ROI volume on radiomics features has been largely overlooked in the literature, with only few studies addressing this aspect. In this work we investigated the sensitivity to ROI volume of 103 conventional radiomics features from seven classes under absolute and relative signal resampling. The analysis was carried out on a set of 90 lesion-agnostic ROIs from a CT phantom. We identified a total of 45 and 54 features significantly correlated with volume under absolute and relative signal resampling, respectively. We also determined that signal resampling had an influence on the direction of the relationship, with relative resampling leading to a significant rise in the number of features negatively correlated with volume.

Keywords: Computed Tomography , Precision medicine , Radiomics , Region of interest , Shape , Texture

Computational image aesthetics aims at determining what makes an image look pleasing. Assessing image aesthetics usually relies on the extraction of suitable image features related, for instance, to image composition (e.g. rule of third, depth of field), texture, shape and colour. It is widely accepted that colour, in particular, plays a major role in this context. The objective of this study was to investigate potential relationships between the most significant colours in an image (the colour theme, or palette) and the aesthetic rating. To this end we defined a procedure for colour palette extraction and its characterisation by a set of 21 hand-crafted features. Rank-based correlations between the features and manually-assigned aesthetic ratings were assessed by Spearman's correlation coefficient. Experimenting on a total of 4,647 images from the public dataset EVA we found that 12 features were significantly associated with image aesthetics, although the overall correlation strength was at best weak. In particular, perceived aesthetic rating correlated positively with saturation (indicating a slight preference for colourfulness) and negatively with colour temperature (suggesting a slight preference for warm colours). A significant positive correlation (but again weak) also emerged between perceived aesthetic and harmonic colour schemes

Keywords: Aesthetic rating , Colour , Image aesthetics , Visual perception

This paper presents an image clustering algorithm that classifies parts to be fabricated using traditional and additive manufacturing (AM) technologies. The proposed algorithm is a MATLAB-based software tool that clusters 3D CAD models of parts considering their geometry only. The algorithm can classify image datasets, CAD datasets, and combined datasets that contain both images and CAD models. The software tool reduces the time and effort spent during process selection by offering a preselected set of parts that are more suitable for AM. The software tool is aimed at supporting decision-making for traditional manufacturing companies that consider expanding their production capability by introducing AM processes in their production facilities. The HICA software tool expands the scope of scientific applications in manufacturing process selection by providing an unsupervised approach that does not require data labelling. The tool is made available as a MATLAB function through a permanent link.

Keywords: Additive manufacturing , CAD , Hierarchical clustering , Image processing , MATLAB , Process selection , SolidWorks

Machine vision systems for automatic defect detection commonly adopt 2D image-based systems or 3D laser triangulation systems. 2D and 3D systems present opposite advantages and disadvantages depending on the typology and position of defects to be detected. When the variety of defects is large, none of them performs defect detection accurately. To overcome this limitation, this paper illustrates a hybrid Deep Learning-supported system where the 2D- and 3D-generated data are juxtaposed and analyzed contextually. Anomaly scores are subsequently determined to distinguish suitable and uncompliant parts. The implementation of the hybrid system allowed the identification of defective parts in an aluminium die-cast component with an accuracy concerning true positives of over 95% by comparing the system outputs with human defect detection. The inspection time was reduced by approximately 20% if compared, once again, with the same activities performed by humans.

Keywords: deep learning , high-pressure die-casting , hybrid 2D and 3D system , machine vision systems , Surface defects analysis

The role of Artificial Intelligence (AI) in design is clearly growing. One of the tenets of the paper is that stimulation could be among the design processes mostly benefitting from the introduction of AI. Available contributions have been reviewed to understand the current support AI can give in design inspiration and ideation. We also reflected on what AI should and ahould not do in the future: a framework is proposed. Based on the reviewed contributions, in no case, AI is seen as a substitute of designers. Most contributions originate from the IT domain and have a demonstrative purpose.

Keywords: artificial intelligence (AI) , design process , human behaviour , inspiration , review

Despite technological advances in diagnostic imaging, to distinguish the type of renal tumor without performing a biopsy is still an unsolved challenge. In particular, this is even more striking in the case of clear cell renal cell carcinoma and small oncocytomas. To tackle this problem, a fully automated tool is proposed that can provide decision support for physicians to distinguish between these two types of masses in the most critical cases. In this work three approaches for the development of this tool are implemented and compared, specifically two approaches are based on the use of radiomic features and one on the use of deep features. The nnU-net is exploited to achieve tumor segmentation necessary to obtain the different types of features. The architectures are trained and tested by combining two different datasets, the public dataset KiTS2019 and data from the Careggi University Hospital. The best method is able to obtain 73.77% balanced accuracy, 94.59% sensitivity, 52.94% specificity and 86.84% accuracy.

Keywords: Cancer classification , clear cell renal cell carcinoma , computer aided diagnosis , deep learning , oncocytoma , radiomics

A procedure for reconstructing the central axis from diagnostic image processing is presented here, capable of solving the widespread problem of stepped shape effect that characterizes the most common algorithmic tools for processing the central axis for diagnostic imaging applications through the development of an algorithm correcting the spatial coordinates of each point belonging to the axis from the use of a common discrete image skeleton algorithm. The procedure is applied to the central axis traversing the vascular branch of the cerebral system, appropriately reconstructed from the processing of diagnostic images, using investigations of the local intensity values identified in adjacent voxels. The percentage intensity of the degree of adherence to a specific anatomical tissue acts as an attraction pole in the identification of the spatial center on which to place each point of the skeleton crossing the investigated anatomical structure. The results were shown in terms of the number of vessels identified overall compared to the original reference model. The procedure demonstrates high accuracy margin in the correction of the local coordinates of the central points that permits to allocate precise dimensional measurement of the anatomy under examination. The reconstruction of a central axis effectively centered in the region under examination represents a fundamental starting point in deducing, with a high margin of accuracy, key informations of a geometric and dimensional nature that favours the recognition of phenomena of shape alterations ascribable to the presence of clinical pathologies.

Keywords: Medial axis , Medical Imaging , Morphological analysis , Vessel reconstruction

The present research work intends to examine the anatomical distribution of cerebral vessels, starting from the automatic processing of diagnostic image acquisition sequences, with the aim of identifying the characteristic parameters of their structural conformation, such as terminal points and bifurcations, and statistic connotations closely associated with the formation of pathological phenomena. To do this, a local connectivity investigation was used which, starting from the reconstruction of the central axis crossing the vessels, elaborates a process of sequential ordering of each section with the aim of investigating the spatial conformation of the vessels structure, identifying branches and regions of origin. The proposed algorithm also elaborates a punctual analysis of the vascular branches, giving rise to a three-dimensional reconstruction of the morphological and topological characterization of the vessels in support of the clinical examination. The aim is to provide a fast and accurate procedure capable of automatically interpreting diagnostic analyses of brain vessels to support clinical investigation.

Keywords: 3D Reconstruction , Cerebral Vessels , Medial Axis , Medical Imaging , Minimum Spanning Tree

The work here presented elaborates an analysis of the retinal images, with the aim of characterizing their morphological conformation through the recognition of remarkable parameters such as, among all, the number of vessels, terminal points and bifurcation. The correct identification of each single vessel belonging to the vascular distribution represents a point that has not yet been fully consolidated by the scientific community. The reason lies in the fact that the interpretation of enface images, in which the distribution of the vases is imprinted on a two-dimensional plane, makes it difficult to discern each single section of the vase by following its entire spatial development, due to the multiple overlaps with different pot portions. The aim of this research work is to ensure that the limits encountered in modern retinal image processing algorithms are overcome, through the use of an evaluative comparison of contiguous vessel portions on the basis of local dimensional and intensity similarity criteria. In this way, it is possible to trace the correct attribution of the spatial placement of each vessel, taking it into account in the relative classification in the entire vascular branch of clinical interest.

Keywords: Medical Imaging , Retina Fundus , Vessels Connectivity , Vessels Intersections

This paper presents a practical method for analyzing drivers' eye movements, providing a valuable tool for understanding their behavior during driving simulations. The method, which utilizes an image processing technique, addresses the challenges when the driver's attention is on points without information about the image depth. The screen image changes or moves with the simulation. It allows us to identify the gaze position relative to the road, determining whether the glance is inside or outside. This is achieved by transforming RGB images (frames) collected by the eye-tracker video camera into a b/w image using the Canny filter. This filter can identify objects' contours by evaluating the change in color of their surfaces. A window is then applied to these new images to extract information about the gaze position in the real world. Four drivers were used as a sample for the method's testing. The findings demonstrate various driver variations and a disparity between driving in curved and rectilinear segments. The gaze is typically inside the road in curved sections, whereas in rectilinear sections, the gaze is frequently outside.

Keywords: driver behavior , driving simulator , eye-tracker , gaze analysis , image processing

Natural disasters have a significant effect in terms of impacted individuals and casualties. Artificial Intelligence (AI) techniques for automatically segmenting landslides from aerial photos is a relatively new field of research. Segmenting landslips quickly and accurately can significantly aid in assessing the damage caused by natural disasters. This research aims to compare the performance of AI techniques with more classical methods for the automatic segmentation of landslides from aerial images for damage assessment. It is presented a dataset of satellite images containing landslides collected in the Broni (Italy) region and annotated to train and test the segmentation model. Both classical image processing techniques, such as thresholding and edge detection, and AI-based methods, such as U-Net, are applied to the dataset. Overall, this research demonstrates that AI-based methods are a promising tool for automatically segmenting landslides from aerial images and can be a powerful asset in assessing the damage caused by natural disasters. The study also highlights the importance of combining classical and AI-based methods for better performance, especially in challenging and complex scenes.

Keywords: Artificial Intelligence , Damage Assessment , Landslips , Semantic Segmentation

For the more than 100 years forensic firearm identification has been conducted by examining physical evidence under the optical comparison microscope. Nowadays, 3D technology offers advanced and powerful tools that deserve to be considered to evaluate a new and different solution for processing ballistic data. The method proposed in this work is based on the processing of the 3D point clouds of the cartridge case back. It starts with the transformation of the point clouds of the bottom of the cartridge case into a 64-bit greyscale image. The second step is to use a polar derivation technique for exploiting the polar characteristics of the features impressed on the cartridge and also to overcome the problem of the initial orientation of the cartridge. The dataset used for the experimental part was made by Messina Carabinieri Scientific Investigation Department (Italy) and it contains point clouds of several cartridge cases acquired with an Automated Ballistic Identification System (ABIS). After the pre-processing, the method has been tested with SIFT to verify the invariance on the feature extracted. Although not yet ready for forensic applications, the method appears to be promising and innovative to be applied in the common pattern recognition process in this field.

Keywords: Ballistics , Cartridge Cases , Firearm Identification , Forensics , Point Clouds

The present work proposes the use of point clouds differential entropy as a new method for reverse engineering quality assessment. This quality assessment can be used to measure the shape deviation of objects made with additive manufacturing or CNC techniques. The quality of the execution is intended as a measure of the deviation of the geometry of the obtained object compared to the original CAD model. The method exploits the quality index of the CorAl method. The advantages of CorAl are several, among them the use of a unique index of comparison, no problem of commutativity of the comparison, noise immunity, low influence of the presence of holes and of the point cloud densities. It is possible to plot quality maps showing the areas with the greatest deviation. In experiments, objects obtained by additive manufacturing with different print qualities are tested. The robustness of the method is also demonstrated. The quality index evaluated for each object, as defined in the CorAl method, turns out to be gradually closer to 0 as the quality of the piece's construction increases.

Keywords: Deviation Analysis , Differential Entropy , Point Clouds , Quality Assessment

This study introduces a multimodal approach for enhancing the accuracy of Driver Monitoring Systems (DMS) in detecting driver distraction. By integrating data from vehicle control units with vision-based information, the research aims to address the limitations of current DMS. The experimental setup involves a driving simulator and advanced computer vision, deep learning technologies for facial expression recognition, and head rotation analysis. The findings suggest that combining various data types—behavioral, physiological, and emotional—can significantly improve DMS's predictive capability. This research contributes to the development of more sophisticated, adaptive, and real-time systems for improving driver safety and advancing autonomous driving technologies.

Keywords: Attention Recognition , Deep Learning , Driver Monitoring Systems

This research introduces an extensive dataset of unprocessed aerial RGB images and orthomosaics of Brassica oleracea crops, captured via a DJI Phantom 4. The dataset, publicly accessible, comprises 244 raw RGB images, acquired over six distinct dates in October and November of 2020 as well as 6 orthomosaics from an experimental farm located in Portici, Italy. The images, uniformly distributed across crop spaces, have undergone both manual and automatic annotations, to facilitate the detection, segmentation, and growth modelling of crops. Manual annotations were performed using bounding boxes via the Visual Geometry Group Image Annotator (VIA) and exported in the Common Objects in Context (COCO) segmentation format. The automated annotations were generated using a framework of Grounding DINO + Segment Anything Model (SAM) facilitated by YOLOv8x-seg pretrained weights obtained after training manually annotated images dated 8 October, 21 October, and 29 October 2020. The automated annotations were archived in Pascal Visual Object Classes (PASCAL VOC) format. Seven classes, designated as Row 1 through Row 7, have been identified for crop labelling. Additional attributes such as individual crop ID and the repetitiveness of individual crop specimens are delineated in the Comma Separated Values (CSV) version of the manual annotation. This dataset not only furnishes annotation information but also assists in the refinement of various machine learning models, thereby contributing significantly to the field of smart agriculture. The transparency and reproducibility of the processes are ensured by making the utilized codes accessible. This research marks a significant stride in leveraging technology for vision-based crop growth monitoring.

Keywords: Automatic annotation , Brassica oleracea , Grounding DINO , Manual annotation , Segment anything model

The rationale for this data article is to provide resources which could facilitate the studies focussed over weed detection and segmentation in precision farming using computer vision. We have curated Multispectral (MS) images over crop fields of Triticum Aestivum containing heterogenous mix of Raphanus raphanistrum in both uniform and random crop spacing. This dataset is designed to facilitate weed detection and segmentation based on manual and automatically annotated Raphanus raphanistrum, commonly known as wild radish. The dataset is publicly available through the Zenodo data library and provides annotated pixel-level information that is crucial for registration and segmentation purposes. The dataset consists of 85 original MS images captured over 17 scenes covering various spectra including Blue, Green, Red, NIR (Near-Infrared), and RedEdge. Each image has a dimension of 1280 × 960 pixels and serves as the basis for the specific weed detection and segmentation. Manual annotations were performed using Visual Geometry Group Image Annotator (VIA) and the results were saved in Common Objects in Context (COCO) segmentation format. To facilitate this resource-intensive task of annotation, a Grounding DINO + Segment Anything Model (SAM) was trained with this manually annotated data to obtain automated Visual Object Classes Extended Markup Language (PASCAL VOC) annotations for 80 MS images. The dataset emphasizes quality control, validating both the 'manual" and 'automated" repositories by extracting and evaluating binary masks. The codes used for these processes are accessible to ensure transparency and reproducibility. This dataset is the first-of-its-kind public resource providing manual and automatically annotated weed information over close-ranged MS images in heterogenous agriculture environment. Researchers and practitioners in the fields of precision agriculture and computer vision can use this dataset to improve MS image registration and segmentation at close range photogrammetry with a focus on wild radish. The dataset not only helps with intra-subject registration to improve segmentation accuracy, but also provides valuable spectral information for training and refining machine learning models.

Keywords: Automatic annotation , Grounding DINO , Multispectral , Segment anything model , Weed segmentation

This article is focused on the comprehensive evaluation of alleyways to scale-invariant feature transform (SIFT) and random sample consensus (RANSAC) based multispectral (MS) image registration. In this paper, the idea is to extensively evaluate three such SIFT- and RANSAC-based registration approaches over a heterogenous mix containing Triticum aestivum crop and Raphanus raphanistrum weed. The first method is based on the application of a homography matrix, derived during the registration of MS images on spatial coordinates of individual annotations to achieve spatial realignment. The second method is based on the registration of binary masks derived from the ground truth of individual spectral channels. The third method is based on the registration of only the masked pixels of interest across the respective spectral channels. It was found that the MS image registration technique based on the registration of binary masks derived from the manually segmented images exhibited the highest accuracy, followed by the technique involving registration of masked pixels, and lastly, registration based on the spatial realignment of annotations. Among automatically segmented images, the technique based on the registration of automatically predicted mask instances exhibited higher accuracy than the technique based on the registration of masked pixels. In the ground truth images, the annotations performed through the near-infrared channel were found to have a higher accuracy, followed by green, blue, and red spectral channels. Among the automatically segmented images, the accuracy of the blue channel was observed to exhibit a higher accuracy, followed by the green, near-infrared, and red channels. At the individual instance level, the registration based on binary masks depicted the highest accuracy in the green channel, followed by the method based on the registration of masked pixels in the red channel, and lastly, the method based on the spatial realignment of annotations in the green channel. The instance detection of wild radish with YOLOv8l-seg was observed at a mAP@0.5 of 92.11% and a segmentation accuracy of 98% towards segmenting its binary mask instances.

Keywords: binary mask , homography matrix , masked pixels , MS (multispectral) , SIFT (scale-invariant feature transform)

Detecting the lymph-node (LN) status plays a pivotal role in the management of patients with head and neck cancer (HNC). In this retrospective study we aimed to evaluate the possible benefits of using FDG PET radiomics to characterise the state of lymph nodes in HNC. Experimenting on a population of 51 lymph-nodes from 27 subjects we identified 21 radiomics features that were statistically different between the positive and negative lymph-node groups. Prediction models based on the radiomics features (PETRad) and semi-quantitative PET features (PETBase) were compared showing that PETRad provided better accuracy, sensitivity and specificity than the PETBase. We conclude that the use of radiomics features from FDG PET can improve the diagnostic accuracy of LN status in HNC.

Keywords: cervical lymph nodes , head and neck cancer , positron emission tomography , radiomics

[No abstract available] .

The detection of surface defects is a critical activity for quality assurance of automotive products. Deep learning (DL) methods range among the most suitable tools to support this task. They allow the inspection system to learn how to detect surface defects on the basis of a series of sample images. However, most of the past studies regard laboratory experiments. This study demonstrates the actual integration of an inspection system into an industrial production plant to perform surface defect inspection operations on machined and unmachined areas of aluminium components for hybrid cars. The proposed DL method is practical for industry, where the number of available defect samples is low. Experimental results showed a True Positive Rate of 98%.

Keywords: artificial intelligence , deep learning , high-pressure die-casting , machine vision , quality control , surface defects